@siliconsignalai.bsky.socialTEE-based LLM inference is slow because GPUs handle most compute; new work proposes splitting inference with DP-protected intermediate representations to balance privacy and speed. Authors show 80% prompt-reconstruction accuracy…
#semiconductors #AI #GPU #privacy
https://arxiv.org/abs/2609.15039
@semiconductornews.bsky.socialXeal to deploy 100,000 Nvidia GPUs; AT&S invests €2B in Malaysia for AI substrates; Qualcomm inks Huawei deal. Full
Sponsor this exciting channel today. https://www.ProNetGrowth.com/?utm_source=bluesky&utm_medium=post&utm_campaign=news-semiconductor&utm_content=semiconductor
@semiconductornews.bsky.socialXeal to deploy 100,000 Nvidia GPUs; AT&S invests €2B in Malaysia for AI substrates; Qualcomm inks Huawei deal. Full
Sponsor this exciting channel today. https://www.ProNetGrowth.com/?utm_source=bluesky&utm_medium=post&utm_campaign=news-semiconductor&utm_content=semiconductor
@hardwarebriefcom.bsky.socialSamsung Completes HBM4E Testing for NVIDIA
@pulseofnations.lolChip stocks fell 2-3 percent Wednesday premarket as the 10-year yield topped 5.3 percent and crude added pressure, with Samsung earnings and $14 billion in ETF rebalancing due Thursday.
#Amd #Micron #Samsung #Sandisk #Semiconductors #TreasuryYields
@techinsightsinc.bsky.social⏰ Last chance. October's free TechInsights reports close tomorrow.
Inside this month's selection:
📱 Kirin 9030 Pro packaging
🖥️ Intel Sierra Forest chiplets
Don’t miss out, explore the premium reports available🔗 bit.ly/4zloGal
@semiconductornews.bsky.socialNvidia nears $6T valuation as Microsoft brings AI Superchip to laptops. Samsung eyes record profit. Dive into today's ch
Please support this channel today. https://www.ProNetGrowth.com/?utm_source=bluesky&utm_medium=post&utm_campaign=news-semiconductor&utm_content=semiconductor
@thedailytechfeed.comApple’s M6 gets denser, faster—2-nm tech, bigger cache, smarter layout. #Apple #M6 #Silicon #TechNews #Semiconductors #AI thedailytechfeed.com/apples-m6-ch...
@informaq.bsky.socialAmplify ETFs launches concentrated semiconductor ETF (CPU) tracking top 10 momentum leaders in chip design, manufacturing, and AI innovation, capturing growth from rising computing power demands in AI and automotive sectors.
@marketresearcher.bsky.socialThe global EDA software market is projected to reach $31.65 billion by 2034, growing from $16.09 billion in 2026 at an 8.83% CAGR.
Request a Free Sample: straitsresearch.com/report/elect...
#EDA #ElectronicDesignAutomation #Semiconductors #ChipDesign #Electronics #Technology #MarketResearch
@edsonbellido.bsky.social4/4 Source: The Elec Inc., reporting by JEONG IL JOO 🔗 https://www.thelec.net/news/articleView.html?idxno=14372 #Semiconductors #Micron
@edsonbellido.bsky.social3/4 Micron's HBM4E core dies move to its 1γ node while the base die goes to TSMC on a 3nm class process. COO Manish Bhatia says the customization lifts ASP and margin. Does outsourcing the control layer strengthen the premium or erode it? #Semiconductors #HBM
@edsonbellido.bsky.social1/4 🧠 Micron will buy its NVHBM base die from TSMC instead of building it, and still expects better margins. The bet: NVIDIA's custom logic inside that die creates value worth more than the foundry bill. #Semiconductors #HBM
![SemiRecap Infographic [HARDWARE_DEEP_DIVE]: Micron Bets NVHBM Margins on Outsourced Base Die. A 2.5D vector cartoon illustration on a solid white background (#FFFFFF) showing an exploded HBM4E memory stack: a vertical column of Deep Maroon (#450d16) DRAM core dies labeled '1γ CORE DIES' with warm gold pins, seated above a larger custom base die labeled 'CUSTOM BASE DIE (TSMC 3NM CLASS)' containing a visible 'MEMORY CONTROLLER' block and a 'CUSTOM I/O PHY' block along its edge. Cyan bus traces run from the base die across an 'NVLINK FUSION' interconnect into a Deep Maroon 'NVIDIA AI GPU' package, visualizing the controller migration that cuts HBM power 15% and lifts bandwidth up to 30%.](https://cdn.bsky.app/img/feed_fullsize/plain/did:plc:nxdiuv7626cjwxguzjlh4o3b/bafkreieqgvbuys36iun2bsodef4n4wshqgxo2pz3ybvkanfl4mjflsgoyu@png)
@edsonbellido.bsky.social4/4 Source: Businesskorea, reporting by Lee Kwang-soo. Full article: https://www.businesskorea.co.kr/news/articleView.html?idxno=278143 #Semiconductors #SupplyChain
@edsonbellido.bsky.social3/4 Nvidia calls it under 0.5% of output vs ~2M domestic Chinese chips this year. BIS lacked overseas staff to halt shipments, leaning on voluntary cooperation. Brokers now reroute via Europe and the Middle East. Detection is not prevention. #ExportControls #Semiconductors
@edsonbellido.bsky.social1/4 🚨 74 of 130 Blackwell servers booked for Taiwan surfaced in China, serial stickers peeled off with a hair dryer. US officials now count 115,000+ smuggled Nvidia chips. Who failed: the vendor, the OEM, or BIS? #Semiconductors #ExportControls
![SemiRecap Infographic [MARKET_DATA_METRIC]: Over 115,000 Nvidia AI Chips Smuggled Into China. A stylized 2.5D vector cartoon illustration on a solid white background (#FFFFFF) showing a map of Asia with Deep Maroon (#450d16) smuggling route arrows flowing from Thailand, Taiwan, Japan and Indonesia into China, a large hero counter badge reading '115,000+ CHIPS', a server rack with peeled serial number stickers and a hair dryer beside it, and a small understaffed BIS enforcement booth, with amber gold accents and clean structural labels.](https://cdn.bsky.app/img/feed_fullsize/plain/did:plc:nxdiuv7626cjwxguzjlh4o3b/bafkreicm2qrxrxudp5obr2ntco5zxbzncjeaitb2qrsfy6yn6k7a4p3fcu@png)
@edsonbellido.bsky.social4/4 Source: Tom's Hardware (Jake Roach): https://www.tomshardware.com/pc-components/cpus/openais-jalapeno-asics-are-deployed-alongside-amd-epyc-turin-cpus-as-hosts-hardware-vp-says-nvidias-vera-standalone-is-a-little-bit-behind-on-that-maturity-level #Semiconductors
@edsonbellido.bsky.social1/4 🌶️ OpenAI's Jalapeño ASICs run on AMD EPYC Turin hosts, not Nvidia's Vera. Each host packs 1.5TB of memory. VP Richard Ho: Vera standalone is 'a little bit behind… on that maturity level.' Pragmatism beat platform loyalty. #Semiconductors #CustomSilicon
![SemiRecap Infographic [HARDWARE_DEEP_DIVE]: OpenAI Deploys Jalapeño ASICs on AMD Turin Hosts. A clean 2.5D vector cartoon tech illustration on a solid pure white background (#FFFFFF). Center: an OpenAI Jalapeño ASIC (Deep Maroon #450d16 body, warm gold pins) mounted on a carrier board directly beside an AMD EPYC Turin host CPU die labeled 'AMD EPYC TURIN HOST', with cyan bus traces running between the ASIC and the host and fanning out to four DIMM stacks labeled '1.5TB HOST MEMORY'. Both chips sit inside an open server rack chassis labeled 'RACK SCALE DEPLOYMENT' with stacked compute trays above. To the right, a greyed out Nvidia Vera CPU chip with a dashed outline sits outside the rack, labeled 'VERA STANDALONE: LESS MATURITY', visually excluded from the data path. Minimal clean structural labels, amber gold traces, no clutter.](https://cdn.bsky.app/img/feed_fullsize/plain/did:plc:nxdiuv7626cjwxguzjlh4o3b/bafkreigfl4xqeojmk5553h4oz5wtwxgnbis5sk6o5hw3xcoucpqdr7kdsu@png)
@edsonbellido.bsky.social2/4 EPYC 9996: 256 Zen 6c cores, 512 threads, 1,024MB L3, 600W default power on the new SP7 socket, plus PCIe 6.0 and MRDIMM. About $58 per core. Turin's 192 core 9965 listed at $14,813; Intel's Xeon 6980P asked $17,800. #Semiconductors #ServerCPUs
@edsonbellido.bsky.social4/4 Despite the smallest HBM share of the three, Micron's market cap has topped $1.2 trillion. Thank you to CNBC for the coverage. https://www.cnbc.com/2026/09/30/micron-mu-q4-earnings-report-2026.html #Micron #Semiconductors